Deal with registration edge cases in UserCleaner

[Splines]

In #647, we introduced the "UserCleaner" that deletes users that haven't logged in for too long. We now noticed that we don't deal with specific edge cases related to user registration. This means that our initial where clause

User.where(last_sign_in_at: ...INACTIVE_USER_THRESHOLD.ago)
        .or(User.where(last_sign_in_at: nil))

(to determine which users are inactive) is not sufficient to handle all cases. The new special cases are described as docstrings in the app/models/user_cleaner.rb.

Furthermore, we also replace last_sign_in_at by current_sign_in_at in this PR. See this Devise code to see the difference. current_sign_in_at always holds the newest value of the last user sign in, so we want to use that instead of last_sign_in_at, which refers to the last sign in before the current_sign_in_at date. Yes, the names could have been chosen better by Devise 😅

Reviewers

I admit that it was way more difficult than I originally thought to arrive at the new statement, which is a lot more complicated than the original one. Nevertheless, I think its worth it to also handle the edge cases, especially in order not to irritate users with deletion mails when they have just registered (and confirmed their account -> note that also in this case last_sign_in_at is still nil).

• As a challenge: please try to write a test that breaks my new inactive_users method. I've done this multiple times to refine the statement and in the current state, I wasn't able to break it anymore. But maybe you can?
• I've replaced multiple occurrences of :user by :confirmed_user in the tests such that they still work. Please make sure that this does not mean I forgot to handle cases where we should actually also test for non-confirmed users.
• I'm sure there are ways to write the where statement in a more performant way, e.g. by resorting to direct SQL or by using some other tricks. I was happy that I arrived at any workable solution for all the edge cases in the first place and didn't measure anything related to performance. This is just to inform you about this.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 54.45%. Comparing base (c00ae31) to head (57dd34f).
Report is 1 commits behind head on dev.

Additional details and impacted files

@@            Coverage Diff             @@
##              dev     #693      +/-   ##
==========================================
+ Coverage   54.40%   54.45%   +0.04%     
==========================================
  Files         160      160              
  Lines        6744     6751       +7     
==========================================
+ Hits         3669     3676       +7     
  Misses       3075     3075              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[fosterfarrell9]

LGTM, but I think we should find a solution concerning the scopes. My suggestion for now would be to place them in the User model, make a comment there, and we will see how to deal with it when we are in the process of refactoring later on.